PI controller capacitor

We have pA-class stuff that's been in the field for decades, in a clean environment. You can always apply a dab of conformal coat over the critical bits if you expect things to be dirty.

No problem for the comparator or op-amp or MCU, either. (Just get the industrial temp range part, not commercial.)

Tim

Stop looking at 'time constant'. Gain is cheap, just make an INTEGRATOR, the only 'time constant' problem is the output voltage hitting the power supply rail. So, instead of feeding a circa 1V signal into the (resistor/capacitor/opamp) integrator, attenuate it (like, down to 20 mV), and integrate THAT with a 1M resistor and 1 uF capacitor, and a low-offset op amp. It'll take 500 seconds to slew by 10V.

Turnon transient is a problem, though: your capacitor might need a relay to reset the capacitor at power-on time, unless you want to wait a LONG time before using the box.

Right, that works. I like to keep the gain the same in the P and I channel, then you have less worry with capacitor windup.... or use a micro as Spehro suggested.

George H.

Yabbut, there is no ideal integrator, unless you're willing to go to the trouble of using a physical process (e.g. voltage --> frequency, phase --> voltage). Consider what happens if your "integrator" has a shitty capacitor (lots of leakage --> low time constant). :-)

One can consider this, in reverse, to come up with a minimum time constant required for a given performance level. Nothing crazy here. :)

Tim

Thirty years ago, it took a few mylar capacitors and about 70M of resistors to make a two-day integration time. Should still be possible. In a PI controller, soakage ought not be an issue, since it should stabilize with zero volts on the capacitor.

But, if I'd had a suitable microprocessor and a dev system thirty years ago, that would have been my heat-treat furnace control brain. To keep drifts down, I actually had to heatsink the op amp (because it had to have the big supply voltage to get the required ramp time).

Yes, did that recently for a temperature controller. One small SOIC micro, instead of a bunch of op-amps and film caps and high value resistors that have to be kept clean, and PWM generator including triangle wave oscillator and comparator.

Doing it with the micro makes tuning it much easier too - no soldering to try each set new of coefficients, and no need for an external chart recorder thingy to plot the response - it spits it out over the serial port. Also, fixing intergator wind-up and stuff like that becomes much easier.

I did add an extra thermistor and hardware window comparator and relay as an emergency over-temperature cutout though - I don't usually trust software to not start a fire if given the opportunity, nor MOSFETs to not fail shorted. The N.O. relay contacts ares in series with the power to the whole box, and have to be shorted with a pushbutton to start the thing up. The window comparator then holds the relay on for sensible temperature values. Too low a value of the NTC thermistor resistance drops out the relay, and also too high a value does (in case the thermistor wiring breaks). The heated object was mounted on a voice coil, so the safety thermistor wires might fatigue eventually.